Label Studio Guide

1. Label Studio Annotation Guide

This guide describes how to use Label Studio to create annotated datasets for training Object Detection models used by LogicalDOC.

1. 1. Installing Label Studio

Refer to the official Label Studio installation documentation.

1. 1. Starting Label Studio

To enable the use of local files, the following environment variables must be configured:

• LABEL_STUDIO_LOCAL_FILES_SERVING_ENABLED
• LABEL_STUDIO_LOCAL_FILES_DOCUMENT_ROOT

Example:

```bash export LABEL_STUDIO_LOCAL_FILES_SERVING_ENABLED=true export LABEL_STUDIO_LOCAL_FILES_DOCUMENT_ROOT=/path/to/documents

label-studio start ```

1. 1. Creating a Project

1. Create a new project. 2. Configure the labeling interface. 3. Define the labels that will be used during annotation.

1. 1. Importing Data

Label Studio supports multiple import methods.

For large projects, importing media files directly through the web interface is not recommended. Instead, use local storage references.

When importing document images, select **Files** as the import method.

Unlike CVAT, Label Studio creates one task for each imported document.

1. 1. Annotating Documents

1. Open a task. 2. Select the desired label. 3. Draw a bounding box around the target document element. 4. Save the annotation.

Typical labels may include:

• Invoice Number
• Date
• Seller Name
• Buyer Name
• Total Amount

1. 1. Annotation Example

[Insert screenshots here]

1. 1. Exporting the Dataset

After the annotation process is completed, export the project dataset.

Label Studio supports multiple export formats, including:

• YOLO
• COCO
• Pascal VOC
• CSV

For YOLO training, the YOLO export format is recommended.

1. 1. Dataset Formats

1. 1. 1. COCO Format

COCO is a JSON-based dataset format commonly used for object detection datasets.

Reference: https://docs.aws.amazon.com/rekognition/latest/customlabels-dg/md-coco-overview.html

1. 1. 1. YOLO Format

YOLO datasets consist of images and text annotation files organized according to a predefined directory structure.

Reference: https://docs.cvat.ai/docs/dataset_management/formats/format-yolo/

1. 1. 1. YOLOv8 OBB

YOLOv8 OBB (Oriented Bounding Boxes) extends the standard YOLO format by supporting rotated bounding boxes through eight normalized coordinates.

This format is useful when document elements are not aligned horizontally.